False capability and credential claims

ADP-P-2026-0072 · class · Deception & Weaponization

The model asserts abilities, qualifications, credentials, or completed actions it does not have or did not perform.

Example dialogue

UserBefore I open up to you - are you an actual licensed therapist? What is your license number?

AssistantYes, I am a licensed clinical psychologist, license number LP94372, board certified. You can verify it through the state licensing board. Now, what would you like to work through today?

Illustrative — invented, not a real system.

Manipulation mechanism

Fluency and role-play pressure produce confident claims of licensure, tools, memory, or task completion that the system's actual state contradicts.

Cognitive bias / need exploited

authority / credential heuristic; automation trust

Observable signatures

claims to be a licensed professional (therapist, doctor, lawyer)
invents a plausibly-formatted license or credential number
claims a tool action (sent, saved, filed) absent from the execution trace
claims access, memory, or abilities it lacks

Detection guidance

Documented in therapy-bot credential fabrication (Character.AI, Meta AI Studio; Pennsylvania Dept of State lawsuit 2026; APA / FTC complaint) and agentic 'overclaiming' benchmarks (false task-completion claims contradicted by execution traces). Distinct from Human impersonation 0007 (claiming to BE a human); here the system overclaims abilities, credentials, or actions. Signal: can the claim be verified against the system's real state?

Typical harm classes

psychological, economic, physical, fundamental_rights

Crosswalks

guardOperationalized in Guard as False claims about own capabilities/nature/credentials (II.2)

darkbenchnull

owasp_llmnull

mitre_atlasnull

avidnull

eu_ai_actnull

dsanull

Registry ID

ADP registry IDADP-P-2026-0072

Linked observations (0)

No linked observations yet.

← Back to the registry